Women’s role of caregiving for under-five children: Implications for dietary diversity and food security in Ghana

Abstract

This study examines the implications of women’s role of caregiving under-five children on household food security. Using two-stage least square instrumental variable estimation, we analyzed cross-sectional data from 305 smallholder farmers from rural areas in Ghana. The results indicate that household characteristics such as size, gender, and marital status also influence household dietary diversity. A government agricultural program that seeks to improve farmers’ income has the potential to improve household dietary diversity. The most important finding is that the women’s role of taking care of under-five children has negative effect on food household dietary diversity and therefore has negative implications for security of the household in rural settings. However, improved income and education mitigate the negative effects of caregiving to children under five on the household food security. A policy implication of the results is that providing nutrition knowledge to women through food security programming is vital to improving household food security and such programmes must also target family planning.

Keywords:

• Under-five children
• dietary diversity
• food security
• endogeneity
• two-stage least squares
• Ghana

1. Introduction

The food utilization component of food security is concerned with people’s consumption of food that meets their dietary needs. It comprises individual-level distribution, consumption of micronutrients, and the physiological ability to use the food (Coates, 2013). Emphasis is placed on adequate intake of essential nutrients to promote good health. The importance of proper nutrition for children for their intellectual and physical development cannot be overemphasized. However, in low-income countries, most rural poor households are poor and members of these households are often malnourished (Frempong & Annim, 2017). Malnourished people in these low-income countries are characterized by monotonous diets of starchy staples like rice, maize, and tubers which are typical for food-insecurity (Thompson & Amoroso, 2014). Lack of dietary diversity may cause severe problems including malnutrition, and some nutrient-deficiency diseases (Darapheak et al., 2013). Compared with men, the effect is more precarious for women and children. The effect of food insecurity is more on under-five children because their energy and nutrient requirements are high due to rapid growth and development (Chege et al., 2016). Caregiving practices such as feeding children, washing their clothing, and other postnatal care also have implications for children growth and health (Marr et al., 2022).

The intake of different food groups implies diversity in both macro- and micronutrients (Hoddinott & Yohannes, 2002). However, providing household members with non-diversified diets and under-nutrition among the poor children imposes greater burden on households in rural areas (De & Chattopadhyay, 2019). The relationship between food security and child nutrition could be more complex, based on the level of the interactions of the socio-cultural, economic, and environmental practices (Frempong & Annim, 2017). Even though government policy can improve child nutrition at the household level, depending on which level is targeted, the transmission mechanism will differ and the desired impact may be achieved at different time intervals.

Various care-giving practices for under-five have been reported in literature including (exclusive) breastfeeding, solid food preparation and feeding, giving medication, preparing children for school, preparing children to sleep, bathing, washing of their clothing, among other postnatal care (Marr et al., 2022; Sarkodie, 2021). In the case of feeding, Caregiving for under-five years requires that these children are provided with good nutritious food compared to what is given to the other household members. However, evidence suggests that in low-income countries, especially poor households, beyond the weaning periods, under-five children are not provided with any “special diet” but the monotonous diet which the entire household depends on (Faiz et al., 2022; Jabeen et al., 2022;). In view of these complexities, the current study seeks to test the hypothesis that taking care of under-five children has implications for the consumption of a diversified diet that leads to better child nutrition.

While caregiving for under-five could be a role for men and women, in Ghana, caregiving practices for under-five are basically a gender role which is performed by women, especially mothers and, in some cases, grandmothers (Aryeetey et al., 2022). Generally feeding under-five aims to provide them with diverse food to ensure they have adequate requirements for both macro- and micronutrients for their normal growth. However, for a poor household that aims to feed under-five with nutritious food and diverse food, there are several opportunity costs, which include worsening food security of the other members of the household. The opportunity cost also includes the mother’s time that could have been spent as farm labor for the production of food crops. Thus, as a result of the gender roles of women, feeding under-five, their labor and resources available farm, and other productive activities for generating food and income generating become limited and have negative implications for food security of the household.

To help mitigate some of the negative consequences of food insecurity of low-income farmers, especially women, the Government of Ghana in 2010 introduced an agricultural marketing programme to help improve the food security of smallholder farmers. The programme was dubbed the National Food Buffer Stock Programme (NAFCO). The programme was to use buffer stock operations of dual-pricing mechanism of buying help farmers sell their farm produce for enhance/better and stable in income to improve their food security. The initiative has been implemented by the National Buffer Stock Company for over (NAFCO) (see details in Abokyi et al. 2022). Women are at the center of this programme as agricultural marketing in Ghana is predominantly women’s role. The role of women in caregiving to under-five and improving food security of the households is of concern to policy-makers. However, how the role of the gender role of women in giving care to under-five children continues to affect their food security via their household dietary diversity is under research in Ghana. Thus, this paper interns to contribute to filling this research gap.

The determinants of food security have been discussed by several studies (see among others Al-Zabir et al., 2020; Kc et al., 2018; Namayengo et al., 2018; Ogundari, 2017). Socioeconomic factors such as income, household size, gender, and education among others have been found to affect household food security. However, women’s caregiving role of under-five children has not been explored much. This role by women is likely to have implications for improving the food and nutrition security of the households. This paper, therefore, contributes to exploring the relationship between women’s caregiving role of feeding under-five children and food security of the household by means of measuring household dietary diversity. The findings from this paper are expected to provide useful insights for guiding maternal-nutritional policy that has implications on enhancing food security and how governments can help in contributing to improving food security in the mist the role of mothers giving care to under-five children.

The rest of the paper is structured as follow: The next section, materials and methods, presents the food security measure, the econometric approach to the analysis and data for the study. Following this is the results and discussion section which presents the estimated econometric results and a discussion of the results. Finally, the conclusion and policy implications of the study are presented.

2. Literature review

The concept of food security has evolved over the years, with definitions spanning individual, household, national, regional, and global levels. Today, at the household level, food security is when a household has the ability to acquire the food needed by its members to be food secure. Household food security refers to a condition that occurs when all household members have sufficient access to nutritious food that would enable a healthy living (Omotoso & Adesina, 2021). Typically, household food security is composed of availability, accessibility, stability, and utilization of food by household members. This study focuses on food security as relating to access and utilization and measuring these by a proxy metric, the dietary diversity score. There is no gold standard measure or metric for measuring food security at the household level as the concept is multidimensional and complex phenomena (Carletto et al., 2013; Jones et al., 2013; Leroy et al., 2015; Tambo & Wunscher, 2017). The reason is that it is difficult to capture all four dimensions of household food security by a single metric (Jones et al., 2013). As such, several metrics exist for measuring food security at the household. An overview of some of the key metrics that are used in measuring food security is presented in Table 1.

Table 1. Overview of common methods for measuring household food security

Method	What it Measures	Dimension
Household Consumption and Expenditure Index (HCEI)	Expenditure on own food production, purchases, and gifts	Access and Utilization
(2)Coping Strategy Index (CSI)	Strategies adopted for coping with hunger	Access
(3)Household Economic Approach (HEA)	The geographical pattern of wealth and assets	Availability
(4)Anthropometry	Anthropometric index	Utilization
(5)Food Consumption Score (FCS)	Food consumption frequency	Utilization
(6)Household Food Security Index	Hunger level	Availability
(7)Household Hunger Scale (HHS)	Hunger level	Availability
(8)Household Food Insecurity Access Scale (HFIAS)	Food shortage and frequency	Access
(9)Latin American and Caribbean Food Security Scale (ELCSA)	Association between food security and malaria	Availability
(10)Household Dietary Diversity Score/Index (HDDI)	Consumption of 12 groups of food items	Utilization and Access

Sources: Leroy et al. (2015) and Jones et al. (2013)

We adopted the household dietary diversity score (HDDS) as a proxy measure of household food security. Household dietary diversity is a qualitative measure of food consumption that reflects household’s access to a variety of foods and also indicates the nutritional adequacy of individuals’ diet (Hou et al., 2021). Dietary diversity thus provides a proxy measure of food access and utilization dimensions of food security of the household. Empirically, the Household Dietary Diversity Score/Index (HDDI), which is the focus of this paper, has used a proxy metric food security by several researchers (Drescher et al., 2009; Kiboi et al., 2017; Morseth et al., 2017; Ren et al., 2019; Zhou et al., 2020).

HDDS is a score of the number of food groups that members of a household consume over a given reference period (Kennedy et al., 2011: Namayengo et al., 2018). Our choice of HDDS as an indicator of food security is that the HDDS concept dwells on the consumption of 12 groups of food items and provides a measure of two key components of food security, viz. utilization and access of such food groups, both from the macro and micro points of view. HDDS is also a measure of socio-economic improvement of households and is generally regarded as an outcome measure of food security for interventions for low-income countries (Huluka et al., 2019; Namayengo et al., 2018). HDD has been used as a proxy of food security by few studies (see among others Mbwana et al., 2016; Wordofa et al., 2020).

2.1. Determinants of dietary diversity as household food security measure

Drivers of HDDS have been studied quite a lot, with key determinants being income, education, marital status among others (Ingutia & Sumelius, 2022; Kiboi et al., 2017; Lovon & Mathiassen, 2014; Kennedy et al., 2010). We provide a brief review of these drivers of food security at the household level.

First, we discuss household income. Household (family) income has been found to be one of the most dominant drivers of household dietary diversity, both theoretically and empirically. Indeed, family income has been found to have positive effect on dietary diversity in both developed and developing countries including China and in Africa (Codjoe et al., 2016; Frempong & Annim, 2017; Hou et al., 2021; Ren et al., 2019). The literature on dietary diversity is consistent of this positive effect of income on dietary diversity (Drescher et al., 2009; Morseth et al., 2017; Ren et al., 2019; Zhou et al., 2020). Empirically, it is reported that increased income among households in urban areas increases the household’s ability to draw their diets from more diversify foods compared to households in rural areas with low incomes who draw their diets from less diversify food (Hou et al., 2021).

Education has also been found to have positive effect on household dietary diversity (Ingutia & Sumelius, 2022; Morseth et al., 2017). Theoretically, increased education is expected to have an enhanced dietary knowledge and shape the food consumption habits of the households. Also, more educated households are more likely to use the internet and other sources such as mobile phones, and other communication methods to access more and relevant nutrition and health knowledge that could guide their food choices and consumptions (Hou et al., 2021).

Other variables that have been reported to influence dietary diversity of households include household’s size, marital status, and gender (Al-Zabir et al., 2020; Kc et al., 2018; Namayengo et al., 2018; Ogundari, 2017). Although several studies have analysed the effect of household size on dietary diversity/food security, the effect has not been conclusive, with some reporting negative association and other reporting positive association. Thus, the effect of household size on dietary diversity has been uncertain. For instance, while Workicho et al. (2016) report positive association, Rajendran et al. (2017) and Thorne-Lyman et al. (2009) found a negative association. For the positive effect, it is argued that large household size provides farmers more labor to cultivate large farms and increase production and diversify their farm activities which could lead to the production of different crops and even livestock. In another breath, the negative association stems from the fact that large families are more likely to face high budget constraints and therefore forgo expensive food items and focus on cheaper carbohydrates.

In the case of a household with the head married, that could also influence diversity of household diets positively as the choice of food could be a joint decision by both the wife and the husband and each person’s interest could be taken on board leading to a diversified diet (Mekuria et al., 2017). The combined knowledge of the wife and the husband could also result in diversify food choices.

Gender has been reported to have a significant influence on dietary diversity with mixed effects. Workicho et al. (2016) reported that because women are responsible for preparing the household food, they are more likely to have more nutrition knowledge than men. Hence, female-headed households are more likely to have diversified diets than male-headed households. It is also argued that male-headed households are more likely to have more resources, including income, compared to female-headed households in developing countries. Thus, male-headed households have the greater ability to acquire expensive and diversify their food choices. We are therefore not certain about the effect of gender on dietary diversity.

2.2. The link between caregiving of under-five and dietary diversity

The caregiving for under-five children by women parent is a cardinal role for women in most developing countries. It is also the case that women parent is also significant player for providing food for the households. Consistent with global practice and culture, smallholder women play multiple roles within the household including parental caregiving to their children and also perform productive role towards improving the dietary diversity of the household, and this is common in Ghana. The caregiving role of women coupled with their economic role (activities) to improve dietary diversity brings stress that may affect the well-being and other aspects of the parent’s life (Ku et al., 2012; Sarkodie, 2021).

The role strain enhancement theory has been used in literature to explain how caregiving to children affects other aspects of parent’s life (Shorey et al., 2022). The role strain theory postulates that individuals experience role strain when attempting to negotiate and balance multiple obligatory roles (Clark Cline, 2010; Goode, 1960). Role strain is the tension and stress which occur within a person as he/she is the sole person performing a specific role and there are also demands of other roles to be perform by the same person (Goffman, 1990). The theory suggests that people have limited time, energy, and resources at their disposal to fulfill different economic, reproductive, and social roles (Goffman, 1990; Gordon et al., 2012; Kalomo & Liao, 2018). Some may have difficulty in fulfilling the additional role of caregiver for their children. The role strain caused by the multiple roles of some women can impact their stress levels and health outcomes, which negatively affects reported well-being (Green-Davis, 2016). Specific roles of women compete for the time and resources with other roles. Hence, women's role of caregiving to under-five children is hypothesized to compete negatively with their role in improving the dietary diversity of the household, hence having a negative effect on household dietary diversity.

A study on women’s empowerment boosts the gains in dietary diversity from agricultural technology adoption in rural Kenya by (Kassie et al., 2020) found that women adoption of agricultural technology improves their dietary diversity score. However, the caregiving role of women often impedes their adoption of agricultural technology due to various competing interests of their time and resources and women often put their care of their under-five children ahead of technology adoption. Thus, the more under-five children a woman has, the lower the likelihood of adopting expensive agricultural technology and hence the negative effect on dietary diversity.

The number of under-five children a household provides care for is therefore likely to take the labor of the household from farming activities and reduce diversification. In addition, caregiving for under-five children requires more financial resources and hence the more under-five children a household has, the more resources that are meant for agricultural production and diversification activities are reduced. The end result is that household’s ability to acquire more diversified food becomes limited. Thus, we hypothesize that the number of under-five children has a negative effect on household dietary diversity.

3. Materials and methods

3.1. Computing the household dietary diversity score (HDDS)

The computation of the HDDS is based on 12–item questionnaire developed under the Food and Nutrition Assistance (FANTA) project. We computed the HDDS by adopting the standardized questionnaire developed and used by Swindale and Bilinsky (2006). Data collection based on the application of the HDDS questionnaire is relatively simple and takes a short time to complete by enumerators and respondents. The HDDS is computed by counting the number of food group item consumed by the household within a reference period, usually 24-hour period.

However, a drawback of the HDDS is that the 24-hour reference period could underestimate the true dietary diversity of food groups that are consumed, but not consumed on a daily basis, are not captured (Vellema et al., 2016). The Kennedy et al. (2011) suggested that the 24-hour reference period does not provide an indication of an individual’s habitual diet. In the Food and Agriculture Organization (FAO) guidelines for measuring household and individual dietary diversity, the FAO suggested that even though they use 24-hour period, a reference period 3–7 days could be used (Kennedy et al., 2011). In a context where seasonality is a key issue in food consumption, the use of the 7-day reference period captures the HDDS better than the 24-hour period (Matita et al., 2022; Koppmair et al., 2017). To address this drawback, we adopted a 7-day reference period. The 7-day reference period has been adopted by few studies examining HDDS to cure the shortcomings of the short 24-hour recall period (Matita et al., 2022; Gupta et al., 2020; Tanimonure et al., 2021),

Another drawback of the HDDS is the seasonality of food consumption (Abizari et al., 2017; Tetens et al., 2003). Abizari et al. (2017), for example, reported that HDDS is affected by the seasonality food consumption in Ghana. Their criticism is that the 24-hour HDDS calculation could within a season this seasonal food consumption pattern in the season could affect HDDS. They recommended that to cure the drawbacks of the issue of seasonality, it is better to measure the HDDS across seasons or use a relatively longer reference period.

Another criticism of the HDDS method is that the food items are not weighted (Zhang et al., 2017). For instance, the consumption of tea and any cereal consumed have equal weight. To correct this downside, we weighted the food groups with weights based on the nutrient density of their macro and micronutrient content. Weighting gives greater importance to foods such as meat and fish, which have higher nutrient density than, for example, sugar. The highest weights are attached to foods with relatively high energy, high protein quality, and micronutrients that can be easily absorbed (Lovon & Mathiassen, 2014). We apply weights based on the Programme (2008) “nutrient density” weighting which range from 0.5 to 4. Nutrient density describes a food group’s quality regarding caloric density, macro- and micronutrient content (Lovon and Mathiassen, 2014; Sibrian & de Fulladolsa, 2017). Table 2 presents the applied weight for various food groups used in our analysis.

Table 2. The food groups and their weights for HDDs calculations

No.	Food group	Weight	Characteristics
1	Meat (beef, poultry, etc.)	4.00	High protein density; easily absorbable micronutrients (no phytates); energy dense; contains fat.
2	Eggs	4.00	High protein density; easily absorbable micronutrients (no phytates); energy dense.
3	Fish and seafood	4.00	Same characteristics as eggs
4	Milk and milk products	4.00	Highest protein quality p, contains micronutrients including vitamin A, energy dense
5	Pulses, legumes (pigeon pea)	3.00	Energy-dense; high amounts of protein but lower quality than meat; contain phytate-bounded micronutrients; low fat density
6	Roots and tubers	2.00	Energy-dense; protein content lower and of poorer quality than legumes; contain micronutrients (phytate-bounded)
7	Cereals	2.00	Same criteria as roots and tubers
8	Vegetables	1.00	Low energy density; low protein density; no fat; high micronutrients density
9	Fruits (mango, papaya, etc.)	1.00	Same characteristics as vegetables
10	Oils and fats	0.50	Energy-dense but usually no micronutrients
11	Sugar and honey	0.50	Energy-dense; Empty calories (calories from foods with little nutrients), stimulant
12	Miscellaneous (tea, coffee)	0.50	Low-calorie density; stimulant
	Total	26.50

Source: Kiboi et al. (2017), Lovon and Mathiassen (2014), and Kennedy et al. (2010). Note: For the unweighted HDDS, each food group was given a score of 1.

Therefore, to measure dietary diversity and estimate the impact caregiving of under-five by mothers on household food security, we used the unweighted HDDS and weighted as two dependent variables as a measure of household food security. The other variables are also used as control variables. Detailed descriptions of the variables used for the estimation are presented in Table 3. In Table 4, both the unweighted and nutrient-density weighted HDDS are provided.

Table 3. Description of the variables

No	Variable	Variable Description	Measurement of variables	Expected Sign
1	Unweighted HDDS	Unweighted household dietary diversity index ranging from 1 to 12	Number of food groups consumed in a week
2	Nutrient-weighted HDDS	Weighted household dietary diversity index ranging from 0.5 to 26.50	Number of nutrient-weighted food groups consumed in a week
3	NAFCO	Buffer stock policy operational area	Dummy (Policy-on = 1: Policy-off = 0)	+
4	Children	Number of under-five children	Number of children under five years as a measure of caregiving	_
5	Gender	Sex of household head	Dummy (male = 1, female = 0)	±
6	Mar	Marital status of the household head	Dummy (Married = 1, unmarried-0)	+
7	Education	Education of household head	Level of education (1 = has no formal education; 2 = has formal education)	+
8	HS	Household size	Number of people in the household	±
9	Income	Household annual income	The amount in Ghana Cedis	+
10	Children X Edu	Interaction of children and education	Interaction of Children and Edu	+
11	Children X Inc	Interaction of children and income	Interaction of the number of children and Inc	+
12	Children X HS	Interaction of children and household size	Interaction of the number of children and HS	±
13	Land	Size of land available for production	Acres	+
14	Irrigation	Access to irrigation service	Dummy (Access = 1, No access = 0)	+

Note: Land and Irrigation are used as instruments

Table 4. The results of food group consumption by the households

No.	Food group	Percentage consumption of food group
		Yes	No
1	Meat (beef, poultry, etc.)	55.2	44.8
2	Eggs	19.1	80.9
3	Fish and seafood	87.6	12.4
4	Milk and milk products	30.8	69.2
5	Pulses, legumes (pigeon pea)	35.3	64.7
6	Roots and tubers	68.3	32.7
7	Cereals	87.7	12.3
8	Vegetables	28.0	72.0
9	Fruits (mango, papaya, etc.)	28.3	71.7
10	Oils and fats	74.4	25.6
11	Sugar and honey	49.3	50.7
12	Miscellaneous (tea, coffee)	34.3	65.7

3.2. Data

Data for the analysis were obtained from the Ministry of Food and Agriculture (MoFA) of Ghana. They were obtained from a survey, conducted in 2016, on Gender and Agriculture from smallholder farmers which included those participating in the NAFCO program and those not participating in the program through multistage stratification and random sampling. Details of the survey can be found on USAID (2016). Baseline Survey on Gender and Agriculture of Selected Communities in Ghana. Final Report. Feed the Future Agriculture Policy Support Project (APSP). In all, a total of 126 respondents in the policy on areas (participants) and 179 in the policy-off areas (non-participants) were interviewed. A reference period of 1 week was adopted.

A summary of the descriptive statistics of the key characteristics of the sampled farmer is presented in Tables 4 and Table 5. The average unweighted HDDS for the policy-off is 4.87 and for the policy-on 6.96 as shown in Table 4. For the nutrient-weighted HDDS, the means are substantially higher, viz. 10.15 and 17.17, respectively. Table 5 furthermore shows that there are substantial differences in income between the policy-off and policy-on regions. That is, GHS 1540.00 versus GHS 4860.00, respectively.¹ The average education level, household size, land size, and the area irrigated are higher in the policy-on area than in the policy-off area. For the average number of children, the proportion of male-headed households the reverse holds.

Table 5. Descriptive statistics

Variable	Policy-off				Policy-on
	Mean	SD.	Min.	Max.	Mean	SD	Min.	Max.
Unweighted HDDS	4.9	1.8	2.0	10.0	7.0	2.1	2.0	11.0
Nutrient-weighted HDDS	10.2	3.0	4.4	17.7	17.2	4.5	7.1	25.6
NAFCO	0.0	0.0	0.0	0.0	1.0	0.0	1.0	1.0
Children	1.9	1.3	0.0	3.0	1.2	1.3	0.0	3.0
Mar	0.8	0.4	0.0	1.0	0.8	0.2	0.0	1.0
Education	1.3	0.5	1.0	2.0	1.5	0.5	1.0	2.0
Gender	0.8	0.4	0.0	1.0	0.7	0.5	0.0	1.0
HS	6.4	2.2	2.0	16.0	6.5	2.6	3.0	17.0
Income (in 1000)	1.5	1.5	0.1	12.0	4.9	4.0	0.1	20.0
Children X Edu	3.6	3.7	0.0	15.0	3.2	3.9	0.0	12.0
Children X Inc	2.9	4.2	0.0	20.7	8.7	6.9	0.0	19.0
Children X HS	12.1	8.8	0.0	39.0	7.7	9.0	0.0	30.0
Land	40.1	62.4	1.0	483.0	171.3	188.6	2.4	1100.0
Irrigation	0.2	0.4	0.0	1.0	0.8	0.4	0.0	1.0

Note: SD is standard deviation, the number of observations is 305; 126 for policy-on and 179 for policy-off

4. Estimation of empirical model

The theoretical model defined in section 2 reads:

(1) $\begin{array}{rl}& LogHD{D}_i={\beta }_0+{\beta }_{1}NAFC{O}_i+{\beta }_{2}Ch{n}_i+{\beta }_{3}Ma{r}_i+{\beta }_{4}Ed{u}_i+{\beta }_{5}Ge{n}_i+{\beta }_{6}H{S}_i+{\beta }_{7}In{c}_i+{\beta }_{8}Ch{n}_{i}Xed{u}_i\\ & +{\beta }_{9}ChnXin{c}_i+{\beta }_{10}ChnXH{S}_i+{\epsilon }_i\end{array}$(1)

where the definitions of the variables are given in Table 2.

In equation (1)(1) $\begin{array}{rl}& LogHD{D}_i={\beta }_0+{\beta }_{1}NAFC{O}_i+{\beta }_{2}Ch{n}_i+{\beta }_{3}Ma{r}_i+{\beta }_{4}Ed{u}_i+{\beta }_{5}Ge{n}_i+{\beta }_{6}H{S}_i+{\beta }_{7}In{c}_i+{\beta }_{8}Ch{n}_{i}Xed{u}_i\\ & +{\beta }_{9}ChnXin{c}_i+{\beta }_{10}ChnXH{S}_i+{\epsilon }_i\end{array}$(1) , there is likely to be simultaneity bias, endogeneity, because of interaction between income and HDDS. That is, income is assumed to affect HDDS and HDDS is likely to affect income. Hence, the explanatory variable income is likely to be endogenous. As a result, the Ordinary Least Squares (OLS) estimator of the model in equation (1)(1) $\begin{array}{rl}& LogHD{D}_i={\beta }_0+{\beta }_{1}NAFC{O}_i+{\beta }_{2}Ch{n}_i+{\beta }_{3}Ma{r}_i+{\beta }_{4}Ed{u}_i+{\beta }_{5}Ge{n}_i+{\beta }_{6}H{S}_i+{\beta }_{7}In{c}_i+{\beta }_{8}Ch{n}_{i}Xed{u}_i\\ & +{\beta }_{9}ChnXin{c}_i+{\beta }_{10}ChnXH{S}_i+{\epsilon }_i\end{array}$(1) is likely to yield biased and inconsistent estimates. To correct for this endogeneity biases, we apply the Two-Stage Least Squares (2SLS) with all the exogenous variables in (1) and two additional variables, land size and access to irrigation, as instruments.

To ensure a parsimonious model, that is, a model that achieves a high level of prediction and a good fit with relatively few explanatory variables, the insignificant variables in the initial model are deleted by a backward stepwise procedure to arrive at the final model (Luo & Ghosal, 2016; Podewski & Weber, 2019). Relatively too many predictor variables may cause overfitting of the model, hence deleting insignificant variables from the model is appropriate (Luo & Ghosal, 2016). Step-wise regression also ensures that the assumptions of normality, linearity, and multicollinearity are not violated (Sletta et al., 2019; Verhofstadt et al., 2016). Hence, in the final model, variables are included only when they are significant in the initial model.

5. Empirical results and discussion

As a first step, we apply the Durbin–Wu–Hausman (DWH) test for endogeneity. The results are presented in Appendix 1. The coefficients of 0.415 and 0.941 for the predicted residual for both unweighted and nutrient-weighted HDDS are significant at the 1% level indicating the existence of endogeneity of income for both the unweighted and nutrient-weighted HDDS-dependent variables. The presence of endogeneity justifies the use of the 2-stage least square approach. As a consequence, we regress income on a set of instruments (section 3) to obtain its predicted value in the first stage of the 2-stage least square estimation. The results of the First-stage 2SLS regression are presented in Table 6.

Table 6. First-stage 2SLS regression

Model (1)
Dependent Variable: Income
Independent Variables	Coefficients.	Std. Err.
NAFCO	0.789**	0.350
Children	0.267	0.355
Gender	0.095	0.320
Mar	0.416	0.388
HS	0.184**	0.081
Education	−0.075	0.334
Children X Edu	−0.134**	0.049
Children X Inc	0.063***	0.026
Children X HS	−0.018	0.049
Land	0.020***	0.001
Irrigation	0.688**	0.324
_cons	−0.894	0.928
Number of observations		305
F stat		61.00***
R-square		0.696
Adjusted R-square		0.685
Root MSE		2.310

***, **, and * represent 1%, 5%, and 10% significance level

We note that the essential condition for an instrument to be valid, i.e., it must be sufficiently correlated with the included endogenous regressors but uncorrelated with the error term, and we test the validity with the Shea’s partial R-square (Shea, 1997). The Shea’s partial R-square takes the intercorrelations among the instruments into account. A small Shea’s partial R-square means the instruments lack sufficient relevance to explain the endogenous variable. A higher value indicates that the instruments are relevant (Baltar, 2014). The Cragg and Donald (1993) is used to evaluate the overall strength of the instruments. The Cragg and Donald statistics is the minimum eigenvalue of the generalized F-statistic from the first-stage regression model. The summary statistics of the first-stage equation show a Shea’s partial R-square of 0.483 indicating that the instruments are relevant. The Cragg and Donald statistics (minimum eigenvalue) of 137.002 is compared to the critical value of the Wald test of 19.193 which is significant at 10%, rejecting the null hypothesis that the instruments are weak. Thus, the result shows that the instruments are not weak and therefore valid, indicating that the variables land and irrigation are appropriate and valid instruments for predicting income.

Table 7 presents the initial model estimated by (biased) OLS and 2SLS. Even though the OLS estimates are biased, they are presented as a robust check. The insignificant variables were deleted by a backward stepwise procedure. The final model is presented in Table 8. The discussion of the results is restricted to the results presented in Table 7.

Table 7. The initial OLS and 2SLS estimates

Model (2)
Dependent Variable:	OLS				2SLS
Log HDDS	Unweighted		Nutrient- weighted		Unweighted		Nutrient -weighted
Independent Variable	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Income^2	0.009*	0.005	0.009**	0.004	0.012*	0.007	0.010*	0.006
NAFCO	0.076*	0.040	0.350***	0.034	0.068*	0.041	0.346***	0.035
Children	−0.111***	0.042	−0.076**	0.036	−0.111***	0.041	−0.076**	0.036
Education	0.190***	0.038	0.098***	0.033	0.184***	0.039	0.094***	0.033
Gender	0.103***	0.038	0.098***	0.033	0.102***	0.038	0.098***	0.032
Marital status	0.186***	0.046	0.404***	0.040	0.184***	0.046	0.402***	0.039
Household size	0.001	0.010	−0.018**	0.008	0.000	0.010	−0.018**	0.008
Children X Inc	0.018***	0.003	0.008***	0.003	0.017***	0.003	0.007***	0.003
Children X Edu	0.004	0.006	0.010***	0.005	0.004	0.006	0.010**	0.005
Children X HS	0.0001	0.006	0.004	0.005	0.0001	0.006	0.004	0.005
Cons	1.170***	0.110	1.886***	0.095	1.178***	0.109	1.891***	0.094
No. of obs.	305		305		305		305
F (Wald chi^2)	35.96***		59.15***		372.21***		612.13***
R-square	0.550		0.668		0.550		0.668
Adjusted R-square	0.535		0.657				-
Root MSE	0.277		0.237		0.272		0.233

***, **, and * represent 1%, 5%, and 10% significance level. HDDS is in logs and, hence, the coefficients are semi-elasticities (% changes)

Table 8. The OLS and 2SLS Estimates with only significant variables

Model (3)
Dependent variable:	OLS				2SLS
Log HDDS	Unweighted		Nutrient-weighted		Unweighted		Nutrient -weighted
Independent Variable	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Income	0.008*	0.005	0.008**	0.004	0.011*	0.007	0.010*	0.006
NAFCO	0.078*	0.040	0.349***	0.034	0.071*	0.041	0.345***	0.035
Children	−0.106***	0.013	−0.052***	0.014	−0.105***	0.013	−0.051***	0.014
Education	0.199***	0.035	0.101***	0.032	0.193***	0.036	0.097***	0.033
Gender	0.103***	0.038	0.098***	0.033	0.102***	0.038	0.098***	0.032
Marital status	0.188***	0.045	0.407***	0.039	0.186***	0.045	0.406***	0.039
Household size^3	-	-	−0.013**	0.006	-	-	−0.013**	0.006
Children X Inc	0.018***	0.003	0.008***	0.003	0.017***	0.003	0.008***	0.003
Children X Edu	-	-	0.009*	0.005	-	-	0.009*	0.005
Cons	1.163***	0.075	1.849***	0.080	1.166***	0.074	1.854***	0.080
No. of obs.	305		305		305		305
F (Wald chi^2)	51.75***		81.37***		371.12***		610.42***
R-square	0.550		0.690		0.549		0.667
Adjusted R-square	0.539		0.680
Root MSE	0.276		2.871		0.272		0.233

***-significant at 1%; **-significant at 5%; and *-significant at 10%

Below, we restrict the rest of the discussion to the 2SLS results in Table 7. The overall goodness-of-fit statistics (Wald Chi² and R-square) for both the unweighted and nutrient-weighted 2SLS models in Tables 7 and 8 indicate an acceptable fit. Also, the overall goodness-of-fit statistics in Table 7 compared to Table 8 is virtually the same, indicating that the insignificant variables were appropriately deleted.

The coefficients of children are −0.105 and −0.051 for unweighted and nutrient-weighted HDDS, respectively, lending support to the hypothesis that taking care of children less than 5 years diverts parents, especially mothers’, time and resources from farm work and thus reduces HDDS. The results are consistent with the findings of Frempong and Annim (2017) for Ghana, Egbounye et al. (2017) for Niger, and Iqbal et al. (2017) for Pakistan. Unweighted HDDS refers to the variety of food groups consumed by farmers, whereas the nutrient-weighted HDDS is calculated based on the nutrient density of food items. The large number of young children means more time is needed to babysit and nanny work by parents. Thus, little time is left to cultivate more varieties of crops other than the staples. Infant care diverts resources from agriculture production to homecare.

For both unweighted and nutrient-weighted HDDS models, the coefficients of the interaction term, children X Inc, are positive and significant. The results show that income mitigates the negative effect of under-five children on household dietary diversity (HDDS). A higher income makes it possible for farmers to enjoy nutrient-weighted food groups, as well as the normal food groups and, are greatly secured in their food needs. Under normal circumstances, more children restrict farmers’ ability to enjoy a variety of food crops but increased income negates this phenomenon. The mitigation effect of income is greater for unweighted than nutrient-weighted as poor farmers are more concerned with food access than quality. Furthermore, the coefficient of the interaction term, children X edu, is positive and significant for the nutrient-weighted. However, for the unweighted HDDS, the impact of children X edu is insignificant. The result shows that education mitigates the negative impact of under-five children years. Educated farmers are better placed to distinguish between food groups and their nutrient characteristics and ensure them in their food choices and diets. General education effect extends to children who have special dietary needs. The quality of nutrients in the diets of children during their early years depends mainly on the behavior and decisions of the parents (Jones, 2014). Since most parents are schooled by visiting community nurses on the importance of nutrition for children, educated farmers in our sample seem to put their knowledge to use. Therefore, education can be a mitigating factor on the negative effect of under-five children years. General education effect extends to children who have special dietary needs.

In Table 8, the coefficients of income are 0.011 and 0.010 for the unweighted and nutrient-weighted HDDS, respectively. Both elasticities are significant at 10% level indicating that household dietary diversity increases with income. The results confirm that Bennett’s law applies to farmers in the study area. This is because increased income results in improved purchasing power and the ability of the household to economically access more diverse foods to meet their food needs. These findings are consistent with the results reported by Taruvinga et al. (2013) in a survey of rural households in the Amatole and Nyandeni districts in the Eastern Cape Province of South Africa. Comparing the income effect of unweighted HDDS and nutrient-weighted HDDS, we find that the effects of income are virtually equal, indicating that farmers taste and preferences do not change with increased income. They increase their consumption of food irrespective of the nutrient content when their income increases. The equal income effect may be that as food prices increase, household food choices and allocation practices change. The nutrient composition of their diets is likely to worsen as households choose food groups limited to cheap staples as reported by Meerman and Aphane (2012).

The coefficient of the NAFCO variable for unweighted HDDS is 0.071 and for nutrient-weighted HDDS 0.345. The high estimated coefficient of NAFCO confirms our earlier hypothesis that NAFCO operations enable farmers to acquire a variety of nutrient-laden foods such as vegetables throughout the year as NAFCO ensures stable income. Since most farmers consume most of the vegetables they produce, the buffer stock operations insulate farmers against losses, provide them with assured income, help increase their yields, and stimulate the expansion of agricultural land to produce a variety of crops including nutrient-laden crops, and livestock (see, Abokyi et al. 2018; Sibhatu et al., 2015). As a result, the effect of NAFCO is marked greater on nutrient-weighted than unweighted HDDS.

Education has a positive effect on both the unweighted and nutrient-weighted HDDS. For instance, the coefficient of education for unweighted HDDS is 0.193 and for nutrient-weighted HDDS 0.097. Both are significant at 1%. The results indicate that education tends to affect both unweighted and nutrient-weighted dietary diversities, with educated farmers being not very savvy with their choice of nutrient-laden foods (Smith, 2004). The results corroborate the findings of Powell et al. (2017) who studied the ethnonutrition knowledge of farmers in East Usambara Mountains of Tanzania and the findings of Murendo et al. (2018) in their study of farmers in rural Zimbabwe. Similar findings have been reported by Kiboi et al. (2017) and Ochieng et al. (2017). The results show a larger differential effect of education on unweighted than nutrient-weighted HDDS. For rural and low-income farmers in Ghana, who generally have limited formal education on nutrition and diet, taking the usual three meals a day is more important to them than the choice associated with variety and nutrients (see, Namayengo et al., 2018; Ntshangase et al., 2018). Unless they are well educated about food variety and food groups, most farmers restrict their choice to readily available food groups (Ruel et al., 2018).

The coefficients of gender for both unweighted and nutrient-weighted HDDS are positive and significant at 1%. The results suggest that male-headed households have a better HDDS compared to female-headed households. Several reasons could account for this result. Women farmers in Ghana often face challenges in accessing fertile land, and are poor compared to their male counterparts (Nyamekye, 2015). As a result, women are less able to produce diversified food. When women farmers double as mothers, it affects their production volumes and the diversity of the crops they cultivate, leading to low dietary diversity (Hitomi Komatsu et al., 2018; Ochieng et al., 2017). These factors account for the effect of gender on nutrient-weighted HDDS and unweighted HDDS. The results are consistent with the findings of Ochieng et al. (2017) in their study dietary diversity among agriculture-dependent households in Tanzania, and Nithya and Bhavani (2018) in rural India. All the authors reported that male-headed households have better dietary diversity score than female-headed ones. The results in Table 7 show that the coefficients of marital status for unweighted and nutrient-weighted HDDS are 0.186 and 0.406, respectively. Both are significant at 1% lending support to the hypothesis that married couples enjoy more nutrient-laden foods compared to their unmarried counterparts. Similar findings were reported by Powell et al. (2017) for Tanzania and by Kaloi et al. ((2005)) for Uganda. A possible reason for this finding is that couples could supplement each other’s effort to provide access to food for the household. Marriage promotes better health (Chung et al., 2015), and encourages partners in food choice and healthy diets because couples often care about each other’s nutrient intake (Stack & Eshleman, 1998). As marriage also involves a combination of two significant incomes, married people can afford more expensive nutritious foods. Therefore, the incorporation of the nutrient characteristics into the weighting could see marital status having a greater effect on the nutrient-weighted HDDS than the unweighted HDDS.

The coefficient of the variable HS is significant at 5% for nutrient-weighted 2SLS estimate but insignificant for the unweighted 2SLS estimates, indicating that household size has an adverse effect on nutrient-weighted HDDS. Household’s demand for food increases with increasing household size; households are likely to prioritize the consumption of staples. This finding corroborates that of Ochieng et al. (2017) about the negative association between household size and dietary diversity in Tanzania. Also, Koppmair et al. (2017) reported similar findings in Malawi. As household size increases, the demand for food also increases. Households that are unable to meet this increased demand for food are more likely to prioritize their carbohydrate needs. This could result in vegetable and other expensive food groups not being met.

6. Limitation and areas for future studies

As a limitation, we note that deriving household food security based on a dietary diversity index alone may not be wholly satisfactory since it does not account for the quantities of the food groups consumed and cultural differences. A new dimension would be to see how the results would differ when the HDDS method captures the quantities of the food groups and the cultural differences of the farmers could be considered in future studies.

7. Conclusions

This paper examined the effect of mothers’ caregiving role of under-five children on household food security measured by the household dietary diversity index (HDDS) in Ghana. This study incorporated nutritional density dimensions into the food security measures to provide information on the consumption levels of specific nutrients. The indexes of each farmer were weighted by the nutritional densities of the food groups. To overcome the endogeneity problem of household income, we used the Two-stage Least Square (2SLS) estimator to address the challenge. Comparing the estimated income effects of OLS and 2SLS methods, we find that the 2SLS income effects are larger in magnitudes than the income effect produced by the OLS estimates. However, except for the income variable, the OLS and 2SLS did not show marked differences for both nutrient-weighted HDDS and weighted HDDS.

The findings indicate that the women’s role of caregiving for under-five children hurts the household’s dietary diversity and hence food security. However, improved income and education mitigate the negative effects of caring for under-five children on the household food security. The results further reveal that participation in buffer stock operations has a positive effect on household dietary diversity, providing evidence that buffer stock operations could impact on farmers’ food security. Other household characteristics such as household size, gender, and marital status of household head also influence household dietary diversity. The results imply that improving the income of farmers such as participation in a marketing program such as buffer stock program has the potential to improve the food security of smallholder farmers’ food security. A policy implication of the results is that providing nutrition knowledge to women through food security programming is vital to improving household food security and such programmes must also target family planning.

Acknowledgements

We are grateful to the Agricultural Policy Support Project (APSP), a USAID-funded project, for supporting the survey for data collection for the Ministry of Food and Agriculture, Ghana. The data collection was funded by the Agricultural Policy Support Project (APSP), a USAID-funded project, for the Ministry of Food and Agriculture, Ghana.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The authors received no direct funding for this research.

Notes on contributors

Emmanuel Abokyi

Emmanuel Abokyi is an (agricultural) economist and impact evaluation specialist. He is currently a senior management consultant at the Ghana Institute of Management and Public Administration (GIMPA), Ghana. He holds a PhD from the University of Groningen, the Netherlands, and Master of Philosophy degree in agricultural economics from the University of Ghana. He also has a BSc degree from the Kwame Nkrumah University of Science and Technology, Kumasi.

Bright Owusu Asante

Bright Owusu Asante Asante is a senior lecturer and agricultural economist at the Kwame Nkrumah University of Science and Technology. He has PhD from the University of New England, Armidale, Australia, and master’s degree from University of Ghana, Legon.

Camillus Abawiera Wongnaa

Camillus Abawiera Wongnaa is a senior lecturer and an agricultural economist at the Kwame Nkrumah University of Science and Technology. He has a PhD from the same university and a master's degree from the University of Ghana, Legon.

Notes

1. GHS is the Ghana Cedis, the currency of Ghana.

2. Income in the 2SLS equations is income predicted based on the first stage regression.

3. Note: The variable children x HS was insignificant for both unweighted and nutrition-weighted HDDS as shown in Table 6 and was dropped through a stepwise backward regression approach. Similarly, HS and children X edu were deleted from the unweighted but not the weighted HDDS models.

Appendix 1:

Endogeneity Test

HDDS Regression with Residual from the Income Regression

Table

Independent Variable: HDDS
	Log unweighted HDDS		Log nutrient-weighted HDDS
Variable	Coeff.	Std. Err.	Coeff.	Std. Err.
Income	0.006	0.007	0.007	0.006
NAFCO	−0.300***	0.099	−0.434***	0.085
Children	−0.205***	0.048	−0.313***	0.041
Gender	0.063*	0.038	0.009	0.033
Household size	−0.073***	0.021	−0.188***	0.018
Education	0.224***	0.040	0.172***	0.035
Children X education	0.056***	0.015	0.133***	0.013
Children X Income	−0.009	0.008	−0.051***	0.006
Children X household size	0.006	0.006	0.020***	0.005
Land	−0.008***	0.002	−0.019***	0.002
Irrigation	−0.146*	0.088	−0.527***	0.075
Residual	0.415***	0.110	0.941***	0.094
Constant	1.495***	0.108	2.687***	0.092
Number of observations		305		305
F		32.22***		52.38***
R-squared		0.570		0.683
Root MSE		0.272		0.233

***, **, and * represent 1%, 5%, and 10% significance level